Predictors of Adolescent Engagement and Outcomes: A cross-sectional study using the Togetherall (formerly Big White Wall) digital mental health platform

Online mental health platforms can improve access to, and use of, mental health support for young people who may find it difficult to engage with face-to-face delivery. We modelled engagement and change in anxiety and depression symptoms in adolescent users of the Togetherall (formerly Big White Wall) anonymous digital mental health peer-support platform. A cross-sectional study assessed online activity data from members of Togetherall in UK adolescents referred from mental health services (N=606). Baseline demographics, depression, anxiety, and usage statistics were assessed. Symptom levels among participants who chose to take validated anxiety and depression measures were measured. And participant characteristics were used to predict engagement. Mean number of logins for adolescent members was higher for older adolescents, and for a longer duration than younger adolescents. Mean number of logins and usage time was higher in female adolescents than males. For the total sample, 47.9% of users accessed more than one course, and 27% accessed at least one self-help resource. Gender and age predicted number of joined courses. Greater accessed self-help materials predicted reduced anxiety symptoms. Members mean baseline symptom levels were: GAD-7 between 13.63 and 14.79; PHQ-9 between 16.8 and 18.58. Data were derived from a naturalistic design and modelling of multiple symptom scores should be interpreted with caution. Findings show that adolescents readily engage with an anonymous online platform for common mental disorder, with scope for tailored pathways for different symptom profiles. Members benefit from engagement with Togetherall materials and courses.

Mental health problems are a leading cause of youth disability ( 1 ) with global estimates of 6.5% for anxiety, 2.5% for depression and 13.4% for mental health problems in young people ( 2 ). Untreated mental disorders negatively affect interpersonal relationships, academic attainment, increasing the risk of adult psychological problems ( 3,4 ). Timely, effective early mental health interventions may improve treatment response and long-term outcomes ( 5,6 ). However, over half of young people do not access or receive psychological treatment ( 7,8,9 ). In the UK, only 25% of those referred to specialist Child and Adolescent Mental Health Services (CAMHS) receive treatment, alongside long waiting times ( 10 ). Despite initiatives demand continues to outstrip capacity ( 11,12 ). The treatment gap may also reflect adolescents' preferences for self-management and stigma-related concerns, which inadvertently create barriers to accessing mental health treatment ( 13,14 ).
E-mental health, encompassing psychological approaches digitally delivered via internet and mobile devices, may promote better mental health in young people who may otherwise not access support. Advantages include wider population reach; flexible and convenient access; privacy and anonymity (reducing stigma); scalability and lower cost delivery ( [15][16][17]. Evidence indicates pervasive use of digital mental health services by young people, ( 18,19 ) with high acceptability, satisfaction ( 20,21 ) and treatment efficacy for youth digital interventions targeting anxiety and depression ( 16,22-25 ); with stronger support for online Cognitive Behavioural Therapies (CBT) relative to other digital interventions ( 26,27,28 ). However, there is significant heterogeneity between studies ( 29 ). Transdiagnostic approaches for young people may be as efficacious as disorder specific interventions in e-mental health research ( 25 ), with large effect sizes compared to no intervention ( 26,29 ), and in providing treatment for, and prevention of common mental health problems ( 29,33, 34 ).
E-mental health interventions incorporating guidance also have greater effectiveness and lower drop-out rates in young people than non-guided interventions ( 25,29,32 ).
Attrition is also a widely reported issue in e-mental health ( 16,33 ). Therefore understanding engagement with digital content may improve intervention adherence ( 34 ). Better engagement with digital interventions predicted improved outcomes for common mental health problems in both young people ( 16, [35][36][37] ) and adults ( 38,39 ).
However, there is heterogeneity in how engagement is operationalized (e.g. usage time, login counts and number of completed modules ( 36,40 ). Furthermore, age 41,42 ) and female gender ( 35,43,44 ) have been associated with better treatment response and greater adherence to e-mental health interventions, whereas findings for baseline symptom severity have been more equivocal ( 3,7,42,44,45,46 ).
One candidate digital mental health approach is Togetherall (formerly Big White Togetherall is unique within mental health technology solutions as members are kept safe within the community via a combination of basic AI, monitoring of online posts, and through continuous moderation by professionally registered mental health practitioners ("Wall Guides").
Moderators have immediate access to senior clinicians. Togetherall refers members at imminent risk to local crisis or emergency . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted August 21, 2021. ; https://doi.org/10.1101/2021.08.20.21262337 doi: medRxiv preprint services internationally when needed. It is therefore a safe, low risk environment for young people to support each other's mental health. Moderation can also be used to address potential overdependency ( 53 ) and emotional contagion in web-based peersupport networks ( 54 ), enhancing the quality and safety of e-mental health support services ( 52 ). Narrative and RCT evidence supports the potential of Togetherall ( 50,55 ), with evidence for significant increases in recovery, and decreased anxiety /depression symptoms after use ( 56 ). Greater use of Togetherall ('BWW') in the study was associated with larger improvements in anxiety and depression outcomes ( 56 ).
The current study explored adolescent members' engagement with the platform and modelled baseline symptoms by age and gender. We investigated anxiety and depression symptom levels in these users of hypothesizing that gender and age would predict usage and symptom outcomes; and females and older participants would display greater use of the platform and subsequent greater reduction in initial symptoms compared to males and younger users. We also hypothesised that higher baseline anxiety and lower baseline depression symptoms would predict greater use of Togetherall and symptom reduction. Greater usage metrics and access of intervention components were expected to predict lower final anxiety and depression symptom scores.

Study Design and Ethics
A cross-sectional design explored use of routinely collected online activity data from Togetherall . Registered members' agreement with the Togetherall (BWW) "Terms of Use" formed semi-passive consent to use of data for the research. Processing of data for research purposes was anonymized according to applicable data protection laws. A data sharing agreement was agreed and the study approved by University of Edinburgh School of Health in Social Science Ethics Committee.

Sample
Data were from Togetherall users referred from UK CAMHS (N=1693). Users were excluded if they were >18 years old (N=678). Excluding logins <60 seconds and participants without symptom measures gave a final sample of N=607. Participants were aged between 16 and 18 years (M=17.3, SD=0.71), predominantly female (78.7%) and White British (84.3%).

Procedure
Data were extracted from a Mini SQL database system and comprised anonymized demographic characteristics (gender, age, ethnicity, referrer); login times and duration; questionnaires taken; guided support courses and self-help pages accessed on Togetherall.
All measures were anonymised self-reports. Data were based on user log-ins to Togetherall from August 2016 to June 2019 . Collected usage metrics were averaged for each participant.
Completed anxiety and depression questionnaires on the website were taken as baseline symptoms. For outcomes, scores were considered from first to last completed anxiety and/or depression questionnaire. Number of questionnaires taken per participant was also recorded.
All participants' data were evaluated for engagement. For treatment outcomes, data were included from participants who completed >1 anxiety (N=200) or depression (N=245) questionnaire ( Figure 1). The majority of the sample (N=606) did not take a clinical test.
Members choose when and whether to take a clinical test depending upon their personal motivation to do so at different points in their individual user journeys.

Intervention
Togetherall is a multimodal web-based platform for managing psychological distress and improving wellbeing. It offers a 6-month renewable membership and free access for certain user groups (e.g. universities or state territories). Materials and programmes are based on self-management techniques, informed by evidence-based practice including CBT ( 47,57,58 ). Registered mental health professionals continuously moderate Togetherall forums and offer guidance to users. Togetherall also sends automatic notifications of online activity to members. Service components include guided courses ("Guided Support"), self-help materials ("Useful Stuff"), peer support forum ("The Community") and platform for creating digital art ("Bricks"). Separate from these services, some organisations also commission online psychotherapy ("Live Therapy"). Togetherall's "Guided support" offers free structured programmes on mental health and general wellbeing, lasting 2 to 8 weeks.
Members can sign up for multiple courses, opt-out and choose when to do activities. Enrolled participants receive weekly course activities, notifications and email prompts and are encouraged to use peer support feature from a dedicated course forum. Self-help materials 7 include psychological and health education and advice on skills development. These materials are organized into 8 categories, including emotional health, life-skills, health and lifestyle. Togetherall further offers voluntary self-monitoring of wellbeing on a large number of validated measures (e.g. depression, anxiety, self-esteem). Members are encouraged to complete routine anxiety and depression measures, at first login (baseline) and throughout completion of Togetherall activities.

Usage measures
Total usage time was measured by combining the duration of an individual's BWW logins (minutes). Logins of < 1 minute were excluded and interactions longer than 1 hour were coded 1 hour to limit the contribution of idle periods towards the total count, consistent with other operationalizations of engagement on digital platforms ( 59 ). Number of logins denoted the total number of times each user accessed Togetherall. Average user time per session (minutes) was defined as an individual's total usage time divided by number of logins. Number of courses taken was derived by summing all "Guided Support" courses users joined. Number of accessed self-help materials was an additive measure of all opened selfhelp materials per participant. Zero counts were assigned to users who did not access Togetherall self-help or guided support courses. Time registered (minutes) was calculated as the time difference between first and last access of BWW for each participant.

Anxiety and Depression
The primary anxiety outcome was the Generalized Anxiety Disorder Screen (GAD-7) ( 60 ) which has established general population reliability and validity ( 61 ), evaluating anxiety over the past 2 weeks. The maximum score is 21, with cut-off points of 5, 10, 15 representing mild, moderate and severe levels of anxiety and a score greater than 10 indicating a diagnosis of generalized anxiety disorder ( 60 ). Depressive symptoms were measured via the Patient Health Questionnaire, PHQ-9 ( 62 ), which has good general population reliability and validity ( 63 ). The maximum score is 27, a score of 10 or greater suggest a diagnosis of major depression and scores of 5, 10, 15 and 20 represent mild, moderate, moderately severe and severe depression symptoms ( 62 ). Internal consistency of the PHQ-9 and GAD-7 were not calculated due to database format.
. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. Individual diagnostic tests indicated multicollinearity and highly correlated predictors between usage time and number of logins (VIF>10.5). Therefore, number of logins was dropped as an outcome predictor. Stepwise linear regressions with 10-fold cross-validations were performed to select optimal models with the least prediction error using "caret" ( 69 ).
. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 21, 2021. ; https://doi.org/10.1101/2021.08.20.21262337 doi: medRxiv preprint

Overview
Visual inspection identified one outlier, which was removed from analyses.
Remaining data remained dispersed, consistent with observational studies of real-world engagement with online services ( 33,59,70 ). To evaluate potential bias in the model estimates, analyses were conducted with influential outliers removed, (Cook' (Table 1). Tukey HSD post-hoc tests demonstrated members identifying as female or "other" gender had significantly higher baseline depression scores than males (F=4.96 (2, 480), p=.007) ( Table 1).  . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted August 21, 2021. ; 1 0

Predictors of number of logins
Due to over-dispersion in the initial Poisson regression, a negative binomial model with a log-link function was fitted to predict number of Togetherall logins. Predictors of interest were age, gender, baseline depression and anxiety, with total time registered (minutes) as the control predictor. Non-parametric tests comparing fitted to the residual values, indicated no significant overdispersion in the negative-binomial model (Ratio=1.32, p=.288). Based on the residual deviance test with a chi-squared distribution (critical cut-off for residual deviance, G 2 =656), a negative binomial regression showed better model fit (G 2 =556.2, DF=598, p=.89), compared to a Poisson regression model with the same predictors (G 2 =2995, DF=589, p<.001). No predictors of interest predicted log number of logins ( Table 3). The control variable of total time registered was significant, showing that expected logins increased as time spent accessing Togetherall increased ( Table 3).

Predictors of accessed self-help and guided support
Negative binomial models were used to estimate frequency of use of self-help and courses joined, using the same set of predictors as the model for number of logins. Negative Goodness of fit metrics, based on the residual deviance test (critical deviance value, G 2 =656), showed appropriate fit for the data from the self-help (G 2 =374.11, DF=598, p=.99), and accessed course models (G 2 =559.9, DF=598, p=.87). The control variable of total time of access to Togetherall was a significant predictor of greater log counts for both self-help and courses activity (Table 3). No other variables significantly predicted accessed self-help. For courses joined, gender and age emerged as significant predictors ( Table 3 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) 1 1

Depression
Outcomes were modelled from users who had completed ≥2 depression measures (N=245). Within this subsample, on average, members spent 109 minutes on Togetherall with a mean of 8 logins (Supplemental Table 1 Stepwise regression with forward selection was applied to identify the best performing model for prediction of final depression scores. Ten-fold cross-validation was used to estimate the average prediction error (RMSE) and the model with best number of significant predictors and lowest RMSE was selected. Number of Logins was dropped before selection due to high correlations with Usage Time. Selection started from a full model of 9 predictors, including age, gender, baseline anxiety and depression, usage time, average session duration, number of accessed courses and self-help and the control predictors number of taken questionnaires and total time being registered. The best performing model included predictors of accessed self-help and courses, baseline depression and anxiety scores, total usage time and average time spent per login (Table 4) (Table 4).

Anxiety
Outcomes for anxiety scores were modelled from users with ≥2 anxiety measures.
Mean usage time for Togetherall in this sub-sample was 117 minutes, with a mean frequency . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted August 21, 2021. ; https://doi.org/10.1101/2021.08.20.21262337 doi: medRxiv preprint of 9 logins (Supplemental Table 2). Within this subsample, 46% (92/200) accessed more than one Togetherall self-help resource and 84% (168/200) joined more than one course. Baseline and final anxiety scores were analyzed with number of taken questionnaires included as covariate. Moderate levels of mean baseline anxiety and depression scores were observed (M=14.64, SD=4.35; M=17.42, SD=6.01, respectively) (Supplemental Table 2). There were no significant differences in baseline depression or anxiety scores between the different age and gender groups. Regression lines indicated a decrease in final anxiety scores associated with higher logins and usage time (

Sensitivity analysis: Complete-case analysis
Sensitivity analyses omitting missing data showed gender significantly predicted greater usage time (Supplemental Table 6) and number of logins (Supplemental Table 7); whereas age was no longer a significant predictor of accessed number of courses (Supplemental Table 7). For final depression scores, average session duration did not significantly predict final depression scores while usage time remained a significant predictor (Supplemental Table 8). The models estimating number of accessed self-help and anxiety remained unchanged (Supplemental Tables 7&8). . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) 1 3

Discussion
The current study explored predictors of engagement and outcomes for CAMHS referred users of the Togethreall digital peer-support platform. Results showed that females accessed significantly more guided courses than males, and subsequently reported lower symptoms.. Findings showed adolescents readily engaged with an anonymous online platform to manage common mental disorders. Baseline symptom scores among adolescents using the service were significan, generally in the moderately severe range for depression and anxiety. The Togetherall platform therefore reaches a group of adolescents with significant levels of morbidity, at diagnostic levels of severity, rather than indicative of a simple expression of distress. However, the data suggests that adolescents sign up to the service before their symptoms become clinically severe.
Most members did not complete more than one clinical test in their journey through the service. For he minority of members who did, our hypotheses about baseline anxiety and depression scores did not predict changes in engagement with Togetherall. Accessing more Togetherall self-help materials predicted reduced final anxiety scores, but no other engagement or demographic variables predicted changes in symptom outcomes.
Total usage time and logins were not significantly predicted by participants' characteristics, suggesting similar overall usage of Togetherall in adolescents regardless of baseline demographics and symptom severity. However, these measures of engagement do not capture the extent to which users meaningfully engage with digital support ( 34,36,40 ). For engagement with specific Togetherall components, female gender predicted greater use of guided courses, consistent with evidence that female adolescents show greater module completion in web-based psychological interventions ( 35 ) and better help-seeking intentions towards e-mental health services ( 74,75 ). This may also reflect variance in preferences for different Togetherall features.
Our sample included a narrower age range than comparable studies of age in digital mental health ( 27,41,42 ), but were similar to existing findings for young people's interactions with e-mental health ( 74,76 ). Contrary to our findings for Togetherall component guided courses, we identified no predictors for accessing Togetherall self-help materials. This contrasts with existing evidence for associations between demographic characteristics and . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted August 21, 2021. ; 1 4 intervention adherence in web-based interventions ( 36 ). The Togetherall self-help feature may thus represent a different modality for e-mental health support. Further, meta-analytic data reports limited influence of patient characteristics on effect sizes of self-help interventions for anxiety and depression in youth ( 77 ) consistent with our results predicting Togetherall self-help material usage.
Baseline anxiety and depression did not predict engagement with Togetherall components, contrasting with existing research linking greater e-mental health adherence to lower depression and higher pre-treatment anxiety ( 46,78 ). However, evidence is equivocal, with evidence for associations in contradictory directions ( 35,37,42,79 ). One possibility is that initial symptom severity may not predict adolescents' Togethereall usage, or the relationship We note number of self-help materials accessed predicted decreased final anxiety scores. These results support evidence that adherence to e-mental health programmes predicts . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted August 21, 2021. ; 1 5 greater symptom improvements in young people ( 16, [35][36][37] ). Thus, our findings suggest engagement with Togetherall self-help components improves anxiety outcomes. In contrast, engagement did not predict decreased depression scores, contrary to previous evidence from digital interventions ( 37,79 but see: 83 ). However, final depression scores were significantly lower for users who accessed more self-help materials in complete-case analyses. This Our sample comprised users referred from UK CAMH-services, rather than most existing e-mental health data from healthy non-referred adolescents ( 3 ). Further, the study addressed a gap in the literature around naturalistic engagement with e-mental health services ( 33 ). We controlled for different durations of Togetherall membership, accounting for their variability and effects on symptom change. Moreover, to control for biases in complete-case analyses ( 84 ) we also modelled imputed baseline data, offering a more precise method of missing data handling ( 46,84,85 ), suitable for observational studies ( 86 ). As with other internet interventions, our data was skewed and with high variability in usage metrics ( 33,59,70 ). Future studies could implement mixed effects models to estimate within-participant variability and missing data ( 46,70 ).
. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted August 21, 2021. ; 1 6 Due to the cross-sectional design we cannot infer causal relationships on Togetherall's effectiveness. Furthermore, the clinical significance of the predicted changes in anxiety and depression is ambiguous considering reliable change indices of 4 points (GAD-7) for anxiety and 6 points (PHQ-9) for depression ( 87 ). It is unclear whether individuals in our sample had diagnoses of anxiety and/or depression or whether symptoms were comorbid with other diagnoses. We also only assessed multiple anxiety and depression measures in a minority of participants, thus measurement bias was present. In future, to better track progress, engagement measures could be analyzed prior to completion of each questionnaire on the platform whilst controlling for the time lapse between completed questionnaires ( 70 ). We were also unable to control for members accessing additional Adherence to e-mental health in naturalistic environments is expected to be even lower than in controlled trials ( 33 ). However, for Togetherall , only 33.5% of our included sample discontinued use after first login -notably less than reported drop-out rates in a number of previous controlled studies of other digital interventions ( 35,88,56 ). Our results indicate Togetherall is accessible and acceptable to young people. Behaviour on the platform can predict symptom change, adding to the literature on the importance of studying engagement with e-mental health ( 33 ). Accessing Togetherall self-help materials predicted decreases in anxiety symptoms. Active engagement with Togetherall components might be more relevant to improved symptom outcomes. However, many adolescents in our sample did not access materials and drop-out rates from guided courses should still be examined. Our findings highlight that engagement with Togetherall materials can facilitate improved wellbeing, whilst also addressing the effects of additional platform features and confounding factors of online behaviour.
. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted August 21, 2021. ; 1 7 This echoes survey data demonstrating students with anxiety symptoms were more likely to use self-help resources whilst those with higher depression symptoms were more likely to access anonymous online chat ( 89 ). Additionally, adolescents who reported their gender as other showed elevated baseline depression, comparably to females and greater than males. This is consistent with survey results of higher depression and anxiety related to prior use of e-mental health services in students reporting "other" gender ( 89 ). Overall, our findings suggest Togetherall may offer a supportive community to address the mental health needs in gender-diverse population and calls for research exploring preferred Togetherall features in male and gender-diverse adolescents. Beyond examining individual predictors, embedding theoretical models in the study of engagement and outcome may reveal how interactions with digital platforms generate behavior change, such as Self-determination theory ( 78,90,91 ). Future studies could use this as a framework linked to Togetherall characteristics to facilitate optimal engagement and its relationship with improved wellbeing. is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted August 21, 2021. ; 1

48.
Cheek . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted August 21, 2021. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted August 21, 2021. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted August 21, 2021. ; https://doi.org/10.1101/2021.08.20.21262337 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted August 21, 2021. ; https://doi.org/10.1101/2021.08.20.21262337 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted August 21, 2021. ;  is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted August 21, 2021. .

CC-BY-NC-ND 4.0 International license
It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted August 21, 2021.  . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted August 21, 2021. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted August 21, 2021. ; https://doi.org/10.1101/2021.08.20.21262337 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.  Table 1 Mean and standard deviations of usage metrics and depression outcomes (n=245) Supplemental Table 2. Means and standard deviations of usage metrics and anxiety outcomes (N=200) Supplemental Table 3: Robust regression coefficient for predicted usage time, with no outliers (n=602) Supplemental Table 4: Negative binomial regression predicting number of logins, and selfhelp materials accessed, no outliers (n=602) Supplemental Table 5. Stepwise regression coefficients for predictors of final depression (n=239) and anxiety scores (n=194), no outliers Supplemental Table 6: Robust regression coefficients for predicted usage time, completecase analyses (n=483) Supplemental Table 7: Negative binomial regression coefficients for predicted number logins, Togetherall activities accessed and Togetherall courses accessed, (Complete-case Analyses; n=483) Supplemental Table 8. Stepwise regression coefficients for predictors of final depression (n=235) and anxiety (n=196) symptom scores, Completer analysis.
. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted August 21, 2021. .

CC-BY-NC-ND 4.0 International license
It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted August 21, 2021.  . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted August 21, 2021. ; . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)
. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 21, 2021. ; https://doi.org/10.1101/2021.08.20.21262337 doi: medRxiv preprint Supplemental . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 21, 2021. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 21, 2021. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 21, 2021. ;